Minutes of the June 2, 2005 US ROD meeting
This was a special meeting dedicated to a discussion of the I2C
errors with Guido and Wim who have been at FNAL for the past few
days and who are about to leave for UCSB.
We still do not know why the CERN setup works but the FNAl (and UCSB)
does not.
The timing marging between clock and data lines is about 1 nsec on
position 6, a bit better on position 2 due to different buss lengths.
On position 4, it is the same as on position 6, not clear why errors
are concentrated on position 6. (Have only looked at even positions
because it is hard to turn the rod over).
It was observed that as expected the margin whne talking
to the AOH increases significantly when the resistance is changed
from 80 ohms to 3.3 KOhms. However, with this resistance we move
the problem over: now there is a problem with the AOH when
talking to other devices (DCU, APV, etc). This is not so
surprising because now the clock signal on the AOH is much slower
and the AOH can interpret the pattern as a START.
The best thing to do would be to make the RCs the same everywhere.
They tried several resistors. It looks like 820 ohm makes the
margin about 4 nsec everywhere. Need more info on this (temperature,
batch variations, radiation). Discussions with Sandro
suggest that batch variation and radiation should not be an issue.
At UCSB they also have GS hybrids. It will be very interesting to
make measurements with those.
Going to 0 ohms only increases the margin from 4 to 5 nsec.
The question of a software patch was brought up, i.e., can we
define some sequences of commands that are not allowed in the system.
This does not seem to be really possible.
Guido also wants to study power supply noise. Maybe this contributes
to the high failure rate.
Further studies of header errors by Wim indicate that it is probably
a software issue.
Tests will continue at FNAL today, then at UCSB tomorrow.